Interactive science sandbox
Materials Graph Lab
Turn atoms in a crystal into a computable graph, send distance-weighted messages through their neighborhoods, and inspect how local structure changes a magnetic-property proxy.
Paper lane
Experimental records + crystal structures + graph neural networks
Read the note
Atoms
Perovskite toy cell
Layer 0
Layer
Sample card
ABO3
- Lattice type
- Cubic perovskite
- Magnetic prior
- mixed-site exchange
- Readout target
- ordering-temperature proxy
Atom inspector
None
- Element
- -
- Coordination
- -
- Message norm
- -
Toy message pass
Layer 0
Node features start from element identity and local spin; edges weight messages by distance inside the selected neighbor radius.
This is an explanatory proxy, not a trained CGCNN and not a materials prediction engine.
How the toy works
This is the intuition layer: it uses simplified structures and transparent math to show how a structure-aware model can preserve physical neighborhood information without claiming first-principles physics.
Physical structure
Atoms keep their identities and positions instead of becoming a flat formula string.
Crystal graph
Neighbor relationships become edges, so local geometry is available to the model.
Spin texture
Toy arrows make magnetic ordering visible before the model compresses the structure.
Message passing
Each atom updates from nearby atoms and distances, layer by layer.
Property readout
The graph is pooled into a crystal-level representation and mapped to a target property.